GFF Podcast
The Global Funding and Financing (GFF) podcast is Clearstream’s podcast series for the funding and financing industry, releasing monthly episodes with senior leaders in the space of secured finance covering all major topics shaping the world of collateral, securities lending, repo and OTC derivates leading up to the 2022 edition of the GFF Summit in Luxembourg. Stay connected with the GFF community across the globe and subscribe to our show. Each of the 30 minutes of lively episodes are hosted by Andrew Keith Walker, Finance and Tech Journalist and Christian Rossler, Senior VP, Securities Lending and Borrowing Products at Clearstream. Legal Information here - https://www.clearstream.com/clearstream-en/imprint-1277756
GFF Podcast
The future of AI in securities lending, repo and market infrastructure
Tune in to the newest episode of the #GFFpodcast featuring Bart Coppens, one of the architects of OSCAR, our AI-based application for collateral management, and Fabrice Tomenko, innovation leader and CEO of Clearstream International S.A. Together with host Andrew Keith Walker they explore the latest #AI fintech innovations and the next generation of AI workplace tools.
And welcome back to the GFF podcast. Yes, can you believe it? It's March already. We are back here in the virtual studio after a fantastic show last month, live at the GFF, and I'm delighted to say that we're having a bit of a reunion, because at the GFF I did the AI panel and we had some fantastic guests and joining us here for that. Later in the show we have Bart Coppens, who is director of IntelliSelect and co-founder there, and a man who is all things AI, or as I like to call him, mr Oscar You'll find out why later and also, of course, fabrice Tomenko, who is the CEO of Clearstream International and the leader in the realm of digital trust at the Deutsche Börse Group and a really interesting conversation to be had there is. Fabrice, of course, was the organizer of the AI panel back at the GFF. But before we come to that, it's time for me to introduce my long-suffering co-host very much the RoboCop to my terminator. It is Mr Christian Rosler. Christian, welcome back.
Speaker 2:Hi Andrew Welcome back.
Speaker 1:Thank you, and yes, christian, you know I am very excited about today's show because just yesterday, april the 29th, my new book, creativity how to Rethink, reimagine and Rework with AI, came out, and obviously it's a big topic for me, and it's not just written by me. Actually, it's also written by Dr Chris Lachlan, who is an expert in product innovation and an AI called Isaac, and he's a named author. On the cover, and, as Chris and I set various challenges for Isaac to complete, we write diaries and essays about that, and then Isaac writes diaries and essays about what it's like to work with humans. It's a bit of a sociological experiment. We had a huge amount of fun and, yeah, check my LinkedIn feed. There'll be a link for it there somewhere. Okay, plug over, christian. When it comes to macroeconomics, when it comes to the transmission of monetary policy through central bank infrastructure and all those complex things, you normally set the big picture and I'm interested to get your view as someone who works outside the world of technology, unlike me. What do you think about AI in market infrastructure?
Speaker 2:Well, ai is definitely for the group, I think, a topic, and AI is going to have a significant impact on the highly regulated financial industry in which we are active. So when I say the group, I say, of course, the Deutsche Börse Group, in which Clearstream Banking is part of, and I think that in the group there is obviously some strategic partners that are currently working on a variety of topics to actually enable the scalable deployment of AI across the group. But I think we really have a first AI based application with Oscar, and I think that's why I'm very happy that today we have on the podcast Bart Coppens, together with Fabrice Tomenko.
Speaker 1:Yes, christian, and it is probably time to dive straight in actually and get our hands oily, as it were, in the world of machine learning and robotic process automation and generative AI, large language models and all those other exciting things that we hear a lot about at conferences like CyBARs and we hear a lot about on LinkedIn. But perhaps we need to pick through the hype and focus in on the real applications here. And who better to explain that to us than Bart Coppens, who is the co-founder, obviously, at IntelliSelect? He's designed and overseen the development of self-learning tools for Oscar, Of course. As well as being a long standing consultant in Luxembourg financial scene, he was, previous to that, a principal director of innovation at Accenture. So, bart, no pressure. Can you explain the different kinds of AI we're talking about here, the many different technologies involved, and give us a sort of a big picture view of the sort of tools you're working with?
Speaker 3:Thanks for that question, Andrew.
Speaker 3:That's a nice challenge to do. To explain the whole field of AI in a nutshell, Two major families that you can identify in there. One is what you would call symbolic AI, logic-driven AI, knowledge-driven AI the type of rules-based AI whereby you let the machine do all of the logical deductions, inferences etc. Based on the truths of the problem area and you let it figure it out by itself A little bit what you would call the higher-order reasoning. You let that be done by the machine. It can do that typically a lot better than humans. Can we start to get illogical in our conclusions as soon as there's three, four variables involved with a couple of values. Most people with a double negation in there, most people are lost. These machines can do logical inferences with hundreds of variables across hundreds of rules and always come up with the exact correct conclusion. That's what we call symbolic, the logical-based.
Speaker 3:Now what's been in the news for the last one or two years now is everything which is more data-driven AI. How can we infer or generalize certain insights from the data that we perceive? We actually talk about machine learning. That whole family in artificial intelligence is called machine learning. It's not as if the machine picks up a book and starts to study, Actually, what the machine does. It generalizes the data or it forms a model that represents the patterns that it sees in the data. As such, we say that the machine has learned, or the system has learned, a way of describing the data in a more general approach. In a very vulgar way of putting it, it's a regression Instead of a regression with two parameters. We're not talking about the large language models with 100 billions of parameters, but in the end it's a regression model that draws that pattern on top of the data. You have a black box of reality in it. You have all the relationships, the laws of nature, but you cannot see it. The only thing that you perceive is the data that it emanates. Now we're trying to guess what those inner relationships are. That's machine learning, Whereas the other type of AI, the more logical part, you could see as a legal text, a contractual text.
Speaker 3:What are the rules in there? What are the constraints in there? What is permissible, what is not permissible? Can we derive conclusions out of that? You need to see it a bit.
Speaker 3:The symbolic one is the law. The data-driven one is whatever the judge decides for each case individually. A little bit of that type of thing. You could see it as the repo agreement versus the repo transactions. In the last two, three years, it's mostly the driver has been the strength of the hardware, the cost of memory, the speed of calculation that has allowed for the data-driven side to pick up incredible new emergent behaviors. That's what you see now with the GenAI, the chat, GPTs, the LLMs. On the other side, the whole field of AI keeps evolving, including the symbolic one, including even more weird ones as a genetic AI and all of those. There's some other things out there. I think, as the industry or in general, as senior management dealing with this topic, you need to keep an eye out beyond the next trend, what is coming at the horizon? It'll be another component of AI that will take speed in one, two, three, five years. You need to make sure that you spread the risk, you make sure you stay on top of all of the topics.
Speaker 1:The non-symbolic AI is much more nuanced and relational, much more driven by mathematics. That's more like the neural networks we've heard about in large language, models that use complex probability math to predict what the next word is going to be in a sentence and help the computer to understand the meaning of what you're asking in natural language. Or, as we like to call it, just ask a question in plain English, correct?
Speaker 3:Those are really data-driven models that form patterns that have emergent behavior. That's a very difficult way of saying that. For example, if you see a group of buffaloes running through the savannah, they don't collide, but every buffalo tries to avoid just his neighbors. But if you see the whole group running over the field, same with the birds. You see birds flying in these flocks through the air. It's as if there is an intention to it, as if they're intentionally creating these beautiful figures. But in the end their rules are very simple Just avoid the bird next to me. Same goes with those type of emergent behaviors. By simply predicting the next word and do that at an incredibly large scale, suddenly you have something that describes texts beautifully, that talks about nuclear science with Shakespearean English. Those are emergent properties. That's the beauty of those type of gen AI or systems that create new creative artifacts. It also is part of the problem in the future, because it might start to show behaviors that are unintentional.
Speaker 2:Can I just jump in there, because I think the image of the swarms I mean that, the birds, I think is like auto-regulated systems. But isn't it that regulation plays a big part in how AI is going to evolve, isn't it?
Speaker 3:Yes, absolutely. I just looked it up and it's quite interesting. A new statistic of Gartner came out and it said that by 2026, more than 25% of any AI-based system will need some level of reasoning and logic embedded in it in order to counter a lot of the unintended problems that you might have with more data-driven systems. So there will be partially regulation to govern it, but also rules from within. If you're going to the domain of ethical AI, there you can typically see three levels. How can we talk about ethical systems? Level one is the people who made it. They do it ethically correct. Then you have everything with regulation, but also social contracts. How well all these people behave.
Speaker 3:The second level is the rules that are embedded in the system itself. Are day ethical? Did I put in a rule that said if you are black, then you do not get along? That system is not ethical. The third level is the system itself, giving a conscious level of ethical thinking. But we do need to be very aware of those things. One of the most famous examples is self-driving cars. It's very cultural to decide who would you run over. Would you kill the driver? The guy who actually bought the car? Would you buy a machine that might kill you in certain moments Because you might run over three kids, and it decides to collide into a wall. Those are ethical questions that these systems will need to start to reply to, and there, obviously, regulation is key to help the developers and those systems to at least have a baseline.
Speaker 1:Fabrice, I want to come to you. You're the CEO of Clearstream International. The last time we spoke, you were on the show, of course, talking about the CCMS the new stock exchange system that's been put in place collateral management system in Canada and so I want to come to you with a question that has been written for you by ChatGPT. So I fed in all my notes and I fed in your LinkedIn profile and various other sets of parameters and said I want to interview Fabrice about AI market infrastructure, and this is the question it came up with for you to start off. I won't keep doing this, by the way, but just to put it to the test because we're talking about it, but let's actually see how it works. The question was Fabrice, in overseeing Clearstream International SA, you're at the helm of transforming market infrastructure with technology. So, beyond the current landscape, what emerging AI technologies do you believe will be transformative for security services in the near future?
Speaker 4:Wow, that's a good question and unfortunately I'm not too sure I have access to the crystal ball called ChatGPT as well, but indeed it's a difficult one. First, I'm not an expert in AI, right, and there have been a lot of hype around the new technologies and before talking about AI, we were discussing the last five years about DLT, new technology which was supposed to disintermediate a lot of the intermediaries in the financial sector, and the reality that we see today, five years or even 10 years after we started to look into this, is there are some solutions that are helping the market, but there is no real disintermediation and, let's say, the market adoption is not that high. At the end of the day, Creating new business model and value is something that is real, but it's a very narrow, basically success If we look at what exists today in production and at a production scale, level, right. If we look at AI, I think it's a little bit different because here, first, ai is touching all the domain, all the activities, much beyond just financial industry. The market adoption is quite high, is significantly higher than anything that we have seen. I think we talk about some statistic about adoption or subscription of users of ChatGPT in five days, which has been huge and therefore it's touching everybody our neighbor or family or kids, etc. So the awareness that we have around this new technology even if some of us do not understand exactly how it works but is there, so it's something that we cannot avoid.
Speaker 4:And then the question comes what do we do with this in the business, in all domain of activity? So market infrastructure, collateral management, securities, lending, credit compliance, all these kinds of things, as if we have to look into this. But we look at this in a different way than maybe the other type of technology, and I think there are two aspects of it. First is looking at making processes more efficient right, so I like a tool to help humans to do their job in the most efficient way. Right, looking for information, for example, helping them to scan data and to highlight some specific alerts or behavior that has been predefined. So these are things that ChatGPT can do, or other things, if you program them correctly or if you give them the right set to learn, right.
Speaker 4:And then the other aspect I will say is try to create value. Right, some business value create new way of working which will help some of our market participants to actively do more business with us, right? And this is basically what we are trying to achieve when we are looking at the different innovation we have created. And together, here with Bart, we were at the initiation of what we call SCAR, which is, at the end of the day, it's a simple between bracket GUI, but which facilitates the work of customer clients by simply making it so easy for them to create a legibility schedule that even, let's say, a kids of 10 years to 10 years old can do, right.
Speaker 4:So and this is the way we want to see this so simplify access to our product. Because when we're talking about financial industry, I mean, some people are looking at us with okay, we need a university degree, we need to be an engineer, we need to be a financial specialist to understand how things are done. And, to be honest, it's less and less the case, right? Because thanks to this new way of looking at things, new technology, it makes it simpler and simpler to understand what we are doing.
Speaker 1:Now there is an interesting question here. Isn't there the stems from that which is talking to the digital trust? Because obviously we're all used to new features coming out from the likes of Google or Microsoft, someone like that, which we download and we use and if there's a bug or if it doesn't work properly, there's updates and sort of incremental changes. And the whole software industry, from a retail perspective, tends to work on the basis of releasing these sort of minimum viable products and then incrementally upgrading them as they go and switching new features on that kind of thing. But I'm guessing that in the world of market infrastructure and building very critical legally contracted relationships between the buy side and sell side or between one collateral manager and another, you're not in the same sort of product cycle, are you? You have to launch something that's absolutely reliable, absolutely rock solid and isn't going to have any failures, because the ramifications of that are very different that level than they are in the Silicon Valley retail sense.
Speaker 4:And indeed let's say, because of, let's say, all DNA and highly regulated industry, we are really concerned about delivering quality and resilience, et cetera.
Speaker 4:This does not mean that we cannot be a drag right, and we have demonstrated this with what I mean again sorry, oscar we have been able to develop and to put this in production quite quickly compared to the normal cycle of, let's say, of development that we used to do in the past.
Speaker 4:It's just a different way of working that we apply, and Bart can also explain this because he's applying the same type of a Scrum-aDrive process for development, validation and testing as well. The good thing is now we have, I guess, tools or software that allows us to do that which maybe a couple of years before we didn't have, and therefore every test need to be scripted and validated by people, but today we have automated tests which can run thousands of tests during the day and we can have, we can reduce the investigation, for example, or the correction to the ones which are not passing the 100% test validation. So we reduce basically quite quickly the area of concentration and attention that our team needs to do. So I will not say that marketing infrastructure are not necessarily not agile. It's a different process thanks to new technology. But in that respect I'm not too sure that AI is contributed that much on that way. It's, I think, other type of processes and technology that is used.
Speaker 1:So, bart, tell us a bit more about Oscar. I mean from unpicking the general problem that it's trying to solve through to how do you actually break that down and apply technologies appropriately to deliver the robustness Unie, but also the agility that Fabrice is talking about?
Speaker 3:Well, having looked at the problem of collateral schedules, it was very much like a parallel from the 1980s, when I started off in the industry and you saw these people, 50-year-old experts, and their expertise lie in I know the code of this and I do a transfer of code 25 and the expertise lies in how they dealt with the systems they were given. But that's not where the true business expertise lies. And in all the older systems of capturing eligibility schedules, I saw more or less the same thing happening. Right, it's about I'm the expert at filling out these Excel files because that's what the supplier is asking of me. So, and you saw clients having operations teams like we're dealing with the Clearstreams and we're dealing with the other triparty agents, and that's the expertise of that team. When the true expertise lies in, how do you describe the risk of the profile of collateral you're gonna take? That's where the true question lies. And so, where you take Oscar, it's a good example of that, combining the type of symbolic, subsymbolic, ai together.
Speaker 3:We're using these language models in order to facilitate the user interface, bringing the system closer to the user so that it's more easier to interact with it. Right, we don't take away the complexity of the matter. It is a complex subject matter, defining collateral baskets. You need to think about what it is that you want. But it shouldn't be difficult. The system should make it easy for you. So that's where we bring in that type of language models to facilitate the user interaction.
Speaker 3:But, as you say, andrew, we are here in a world where it then needs to be 100% correct. We cannot have 99% probability that these T-bills will be eligible or not in some cases. It needs to be contractually correct and that's where we bring in that logical part of AI as well to start reasoning with those definitions to see does it make sense If you say that in one rule bombs are eligible and in the other one you say but every single subtype of bomb is not, then that might be correct formulated rules, but they don't make a lot of sense from a business point of view. That's the type of stuff that those reasoning and just can then pick up right. So I think, oscar is a very good example of those what we call Neurosymbolic type systems, where you bring data-driven and logic-driven AI together to actually make complex subject matters a lot easier to handle, bring the machine closer to the human so that he can actually do the true value add and be more effective in executing the job.
Speaker 4:Obviously, when we are using, as a marketing frustrator, this type of tools that we want today to give access to our clients, it's really important that the result is the one expected by the clients, right, and the result that we want as well.
Speaker 4:So the point that Bart is making is very, very important, because there are a lot of discussion today about, and sometimes we are, let's say, looking too much in, generative AI.
Speaker 4:But the result that we need to get from those tools need to be 100% correct, and I know that there are a lot of work which is currently done around this to make sure that each time you ask the same question or you ask the system to do something for you, you get the same result, and it sounds a little bit, let's say, maybe difficult to understand. How is it possible we can get different answer? Bart can explain this much better than me, but the reality is, when we are touching this type of technology, sometimes it's a black box, because there are so many combination, that or path that the system can take in order to get to the result and a small change in the parameter can influence a different response and we cannot really check or, basically, it's not possible for a human to follow all the path, and therefore we need to ensure that the technology reaches certain maturity right, given the comfort to the users that what is the response is the correct one. This is what I was expecting, right.
Speaker 3:These systems, the more Gen. Ai systems are there for being creative and they inherently include some level of randomness in the answer they formulate, so that you ask 10 times the exact same question generate a picture of two people eating ice cream on the beach. Each time you will get a different looking picture, which is fine for that use case. But if you say, generate me 10 times most likely eligible collateral, then you would 10 times get a different set of securities right. Four out of 10 times might contain positions that are not eligible. That is the creativity you do not expect the system to have. Then we come back to how can you make sure that people trust the system to do what they're asking it to do? Sometimes you do want to have the same answer for the same question, Because that is the factual correct question.
Speaker 1:Of course we're talking here about having auditable results where you can very clearly prove well, this is the flow of data and information. Does that make it difficult to how do you surface that data? Because these systems are incredibly complex as they get more complex. How can you deliver that level of audit potentially into a client who maybe doesn't have the in-house skills or the resources to be able to fully digest the ramifications of the system and the data management and different disciplines required to be able to do that audit?
Speaker 3:Yeah, there you have the techniques of validation and verification. As I said earlier, some systems you need to be able to verify the answer immediately. Is this a picture that I like? Is this a proposed email text that I like? You're verifying it on the fly. Other systems need more, deeper validation. That's why Oscar, at this moment, includes what we call a one-step validation of understanding step. Oscar will never react to the pure natural language that you give it. It will always reformulate it in something it accepts, understands and asks you to confirm if he understood it correctly, because as of then, it can reply with 100% certainty to the question. Part of the research that we're doing already in our labs is actually can we move towards multi-step validation of understanding so that the system knows it didn't understand it, but it will then ask you to have relevant question in order to get a better understanding until you confirm that it understood the question? That is the type of user interface, user system interface or dialogue that is quite useful.
Speaker 1:And Fabrice. Coming back to you, I have to say you must, though, be looking at this and thinking. After all the conversations and sessions they had at CyBOS about generative AI, about having artificially intelligent assistants operating not just with retail clients, but operating as assistants in offices to make people more productive, there must be some sort of pressure on you to look at the applications for that slightly more Alexa generative end of the AI spectrum. So what are your views on that? Do you think it's going to be the age of assistants it has been called? Are we going to be living in a world of AI assistants in the office any time soon?
Speaker 4:Let's say maybe two sides or two points on this. The first one, I will say yes, definitely, and this will not be only in office, it will be in our life of day to day. Even with your doctor, you will be facing this type of assistance going forward. So definitely this will happen. Now, I think the pressure in the industry is always to be transparent. Is there something that we can? Sorry, I repeat. So the pressure in the industry is mainly to look at what is the added value that we can make out of this. Can we make more money to a certain extent, even by reducing costs or creating new product? Now, the reality is, we should avoid the trap of looking at this technology and try to implement it for everything that we have a problem with, because, I think Bart explained, you need to apply the right technique, the right combination of tools that you have, to the right problem.
Speaker 4:Is the narrative AI a valid tool for some of the problems we have in the market infrastructure? Maybe not. If we are talking about production activities, it's maybe something else that we need from AI. Natural language processing is definitely something that we see value, and we can extend this Create images. I know it's an easy example, but this is maybe not what we need. It's certainly useful in other areas marketing, for example but not necessarily for what we do with our clients. So I think we need to be careful about not not getting too enthusiastic of using this technology at all costs, because at the end of the day, it will not help. So I think this is the way we are looking at this.
Speaker 1:Christian, I want to come to you and ask about central banks and that sort of public sector finance zone. I mean, what are they saying about it there? I can't imagine there's a lot of use for AI in debt management offices or in the ECB right now, but presumably they are looking at these topics too. What's the mood like for that side of finance?
Speaker 2:Well, that's a big question.
Speaker 2:I think that I haven't certainly seen any AI application.
Speaker 2:When I interface with official public institutions, I think I saw how risk adverse and how cautious they are when it comes to and that's what Fabrice mentioned earlier when it came to DLT and blockchain, which is about this intermediation, and I think AI is definitely a totally different ballgame, as I understand now from listening to Bart and also Fabrice and central banks, central counterparties, central securities depositors, like we are, are central to a market and that's why we call us market infrastructure providers, so we have a central role to play.
Speaker 2:Of course, when it came to this intermediation, the business model itself was at stake, because I mean, as you know, blockchain and DLT and it was mentioned, I think, the word trust, I think when it first came out, there was an article in the Economist which called it a trust machine, so it's actually to put trust in a decentralized system. I mean here we're going next level. So I, to answer your question, I mean I haven't seen any use case for the time being within that official public institutions would when it comes to AI. But I mean, that's my humble opinion.
Speaker 1:Now let's address the AI elephant in the room, of course, which is privacy, client data confidentiality, gdpr. There are many, many different layers, heavily regulated, legally enforceable issues to do with privacy, to do with systems resilience. Obviously, the Digital Operations and Resilience Act is coming into effect in the EU, dora, and I'm sure when we start to explore that, we'll find out that there are more issues to do with using generative AI, because, of course, if you're using something like an engine that's provided by an external supplier that's using a large language model, you can't quite be sure how secure your data and the data sets that you put into it are going to be. So I think we should approach that, because there is a genuine risk here. Isn't there when you're dealing with confidential trading information or financial information for a large client? I mean, fabrice, how do you approach that? Because that surely sits at the core of the digital trust issue when it comes to AI.
Speaker 4:Yeah, and indeed I think this is one of the main issues we are seeing is how do we get to the private privacy of the data if we want some of those algorithms to learn from those data?
Speaker 4:And, obviously, companies you cannot put all the data of the customer through GPD and prompt it. I think there are some solutions. We can see that some of the provider, like Microsoft, is providing access to their AI engine but making sure that there is a kind of privacy and that the data is not, let's say, used outside. But the reality is, depending on what you are looking for, the quality of the answer will depend also on the data that is used for training. So, if you are just doing this in a small sample, what would be really the accuracy of the result? Because it's most probably better to have a bigger sample to run against. So this will be, I think, topics that I'm not sure we have answered to for all the cases, but definitely something that will be at the core of the assessment when we are looking at the different tools and technology that we are to apply to a specific problem.
Speaker 1:But at the panel event, stephen Granger from SWIFT, who is leading their innovation technology team. He talked about federated AI models where you can kind of divide up the data so that belongs to the client, but the processes can live in a centralized, trusted place, for example like SWIFT, so you might have information that could be going to and from an AI that is doing the processing and the clever part, but your data remains yours and separate and never gets revealed. Is that something we should expect to see evolving, or is it going to be more along the lines of slightly more creative or perhaps wacky ideas like modeling, synthetic data, where you use generative AI to completely anonymize everything and then it doesn't matter because the data isn't real anyway. It's just showing the same relationships or numerical or mathematical values, but it's got no identifiable information. This feels like a minefield to help us get through it.
Speaker 3:It'll be a little bit of both. What these new AI models so these data-driven huge gen AI or transformer models need is humongous amounts of data. So you have to be able to analyze it all at the same time to see the high-level pattern and to get that emergent behavior. But that should not necessarily mean that you hand over each of the individuals, that data to each other. So you have, at this point, some governments are actually starting to create data infrastructures like neutral government platforms that allow you to bring data together. They create the model, hand over the model and everybody gets their own data and the centralized trust to the hub destroys it again.
Speaker 3:You see that for certain hospitals doing that, in order for cancer detection on images, they would share the patient information, create the high-level model and then bring that back to all of the different hospitals, but then actually everybody only uses the medical information of their patients. So you could see similar issues coming up and then Swift or Market Infrastructure Plays would be dream places to do those type of services. Create higher-level models with trusted third parties, with neutral third parties in place, because then you need to see a little bit, at what time does data become not personal anymore? If I say the average repo volume is I don't know one trillion, but it turns out there's only one client doing repo. Well, that was not very anonymous, because I know who that is. If you have 300 clients, there is no personal data in that statistic. So there's also types of derived data and the quality of derived data that can already hide a lot of the personal information behind it and still generate quite useful data services but useful patterns and models on the back of that.
Speaker 1:Now it's usually at this point in the show that we start asking our guests the crystal ball questions, because we love to make some predictions about the future. It seems to be that the environment is primed for a big tech jump not a big bang so much as a big surge in innovation and fintech. What do you think, looking into the crystal ball, we're on the verge of a big change in the global funding and financing industry.
Speaker 3:I think if you look at the onboarding, you have the big KYC problem know your customer. There's a lot of tools, techniques, technology being put in place to show who you are and to understand who your counterpart is. Kyc is one thing. The blind spot that I think is still there is become your customer BYC. How do you enroll that client onto the services that you have? Knowing your customer these days is a matter of weeks. Getting him to buy the actual service can take months to put in place. So it's the integration and activation and of the service for your client that is still problematic, not at the least signing all the agreements, agreeing all the back and forth of the negotiation. So that is where I think AI will start to play a crucial role in facilitating that. So authentication techniques for KYC, a lot more logic and knowledge-driven AI to facilitate the contracting process like an Oscar does. It brings the typical months negotiation of a profile down to mere hours. It'll be applicable to a lot more cases as well. And then you talk about the T plus one.
Speaker 3:Then we will definitely go into more of the data-driven or the patterns analysis, and there I would look at the timeframe for financing is now reducing even more, almost to the limit that we're talking about the hour difference across the globe, where you have different regions only having a matter of hours to find the funding for the trading they have been doing. So there's less margin for error, a lot more need for insight the moment that you need it. What is the problem? People don't do repos for fun, right? They do it to get cash. What are the reasons they need that cash?
Speaker 3:Where? What are alternatives to get that cash? How does that report trade in a business case manner compared to other ways of finding cash? And it's getting those insights and those answers in the short time frames when they're needed in a way that you avoid mistakes, errors. That's when you're gonna need the pattern analysis as well as the logic to give you the true, proper answers. And then we're talking about really inside generation, deep inside, together with recommendation engines at a hyper-personalization level that will drive that type of innovation. So that that my two cents on the crystal ball, fabrice what about you?
Speaker 1:I mean, there must be things on the horizon, inflection points of the industry coming where we need all those solutions that Bart talked about, and also, of course, let's not forget, the ever-increasing demand from regulatory authorities, national competent authorities, for more and more fields of data to identify systemic risk in trades, in markets, and that's not just, obviously, in the EU, but regimes from the US across the EU, across Asia. I mean, what's your view? What do you think is gonna happen? Are we at the beginning of a new AI industrial revolution for finance?
Speaker 4:My view is that AI is certainly the fourth revolution, but not necessarily for our industry. It will. As I said at the beginning, this will impact our life on a day-to-day activity and we will see a big change on how we use this tool. You know day-to-day life. Now, if I come to market infrastructure and making a change on the way we are doing things, I'm not too sure.
Speaker 4:The reason I'm saying that is T plus one or T plus zero is not an AI, but right. But what will be necessary in order for us to be able to meet these objectives is we will have to use tools to be more efficient, in order to apply more stricter or simply to keep the pace, because everything will be quicker and we will need to be able to react more quickly than in the past. So if you are thinking about applying regulation, compliance checks, sanctions checks as well, or you provide credit to your clients on a T plus one basis, this will require much more efficient tool. And this is actually where AI right, with the right toolkit not necessarily gen AI in this instance but will help us to cope and to meet with these objectives. So I will say AI will not be necessary. The big foundation for the transformation of capital market and market infrastructure, but it will be definitely tools to be used to support this change.
Speaker 1:At the panel we did at the GFF.
Speaker 1:One of the suggestions that came out from Jorge Sanz, ibm, was we should expect to see these tools, especially generative AI, playing a big role in the way we publish information not just about funds, but the way that we present information, the way that marketing departments produce collateral, as well as the way we publish reports, and there's a lot of activities where these tools actually can be very effective. And we've all met someone who said oh, I get chat GPT to write my emails or, in my case, as someone who spends their life writing, I need a proofreader. That's better than Grammarly and chat GPT is fantastic for actually proofreading and identifying those sort of word blindness errors, because, let's face it, no one can read their own articles and spot all the typos At least I'm hoping no one can, it's not just me. So what about that? I mean, when you look at your sort of operational teams, you look at your marketing departments, do you see that AI is probably going to appear first as a tool to boost productivity and to boost output for those teams first?
Speaker 4:They definitely boost productivity and creativity to some extent, right. Even if we are, let's say, looking at the same data. It's helped people to create new things, right, Simply because of asking maybe we were talking about the two people on the beach eating an ice cream, but this is where the creativity can be boosted, right. What we need to avoid is, to a certain extent, not to fall in the trap of standardization and information, because then all the marketing areas of the different banks will have the same photo to be used, and this is an example, but definitely this will help.
Speaker 3:For our marketing. We're actually not because we're small and facing such a large clients. It's more of a relationship-based discussions and talking. So I don't think that by creating a brochure, intelliselect would have entered at a clear stream. We're talking about networking and relationships over time. I might let my emails be proofread by chatGPT, the same as I have automated spellings check being done on it.
Speaker 3:But I don't think that the likes of Brice would appreciate it if on the reach email I say generated by chatGPT, because how special would my clients feel and I think the same goes for every relationship manager at Clearstream towards their investment bank client, and so they're fund clients, right. You need to be very wary if you say like we're heavily using in all our personal communication GenAI to talk to our highly respected clients, to be careful how my jeopardize some of your relationships, right, but for me, using chatGPT or those language models is just the next evolution of the spellings check. It'll help you and create better sounding sentences. It might get some of the international English mistakes out of there. That's what it's going to do and, as Fabrice said, it'll help in boosting creativity with teams that need creativity in their artifacts that they make, and that's where they're going to be very useful.
Speaker 1:So the final question, of course, as we draw these threads together, would be how long do you think it's going to be before Christian can say hey, chatgpt, what are my collateral liquidity requirements for today's trading? Do you think that's five years off, 10 years off, or is that never going to arrive?
Speaker 4:I don't know, but I can imagine that at a certain point in time, not necessarily a chatGPT, but let's say a more, let's say Alexia type interface will be integrated in the future in the user interface of market infrastructure. This, I'm sure, because we will evolve. As I said before, we are not addressing our services to the same people than before. So they grow up with new technology and this will be the consequence. The usage of more user-friendly tools will definitely be there, and voice control will certainly play an important role in that respect. Now, combined with a chatGPT, a gen-EI, I don't know. And Bob, what about?
Speaker 1:yourself how long before Christian and I really can be replaced by robots.
Speaker 3:That will be never. The two of you are so unique there's no robot that could replace you. But, picking up on what Fabrice said, it's the culture that determines the interface. I still remember when we had big discussions if infrastructure applications would ever be on an iPad, because who would ever use it? I was like well, eventually, if it's being used for everything else in people's lives, it's going to be used for doing those type of things as well.
Speaker 3:Same way, if people now these days are using text and chat and are using more and more voice, it will happen, and it's a matter of infrastructure. The market infrastructure will be the first one to offer that. Maybe, maybe not, but they will definitely end up offering that type of stuff. And so, for the example that you gave, I can immediately find for every step in that problem, solutions that exist. It's just about a matter of time when it is decided that you need to follow up on that and to be maybe a little bit dirty on the internal secrets. It's probably when the former interface applications have been written off that there'll be investment money to do the next one.
Speaker 1:So I guess all that remains is to thank our very special guests this week, and that is obviously co-founder of IntelliSelect, all Things AI and Mr Oscar himself. A huge thank you to Bart Coppens. Bart, thank you.
Speaker 3:It was my pleasure to be here. Thank you, andrew. Thank you Christian.
Speaker 1:And also, of course, a big thank you to the CEO of Clearstream International, the man who put together the AI agenda for the GFF, and it was a great panel, by the way. I mean, obviously, I would say that I was moderating it, but it was for me hugely enjoyable to be on the stage with people who knew that much about the topic. Very entertaining chat. So, fabrice, thanks for that and thanks for joining us today. Thank you very much, andrew. Okay, and Christian, that's it, I guess all the remains.
Speaker 1:Earlier, I said you were the Robocop to my Terminator, but actually that's not strictly true, because the Terminator is ruthlessly efficient and highly advanced and clearly I'm not. So, whereas you are certainly part man, part machine and all securities lending, I'm more like Ed 209, the rather clunky, haphazard robot who is really there for comic relief in the classic Robocop movies and, if you're listening, alexander Rogues, who runs the securities desk there in London. Another movie reference, just for you. He, like Me, is a massive movie buff and on that front, we do definitely owe a big thank you to Mr Christian Rosler, who is the Vice President at Collateral Liquidity and Lending Solutions and a specialist when it comes to developing business with central banks and central bank infrastructure. Christian Rosler, thank you very much.
Speaker 2:Thank you. I would like to thank Barton and also Fabrice for unistifying AI today. I think I really like the information that Bart gave, and also Fabrice. I think it's a great show and I'd like to thank them for being on the show.
Speaker 1:That's great. Okay, Well, Christian, you and I, we're going to be back next month with another one the last two episodes coming up of season three of the GFF and we have some very hot topics coming, don't we?
Speaker 2:Yeah, we are going to run the show on different topics. Obviously, security is landing ETFs back to basics, if I may say so, and then we're going to have a show which is going to be on central bank money for a tri-party report transactions.
Speaker 1:Okay, so that's two big GFF topics there, and I'm very excited about ETFs because obviously, when we talk about technology, etfs have been around a long time. But of course, the big tech news over the last couple of months has been the signing off by the SEC of Bitcoin-based ETFs and crypto ETFs coming, so there's a huge amount of innovation. Potentially, you can find its way into the securities lending markets there and, in the meantime, if you want to network with the real, human version of myself and Christian and Fabrice and Bart, do join us on our LinkedIn page that is linkedincom Slash companies, slash Clearstream, and you'll find out all about the podcast, all the links to all the previous shows, there are clips and follow-ups from the GFF Summit and lots of interesting content. So come and join us there and, in the meantime, from everyone here in the virtual studio, from everyone here at Clearstream and, of course, from myself and Christian, have a good month, have a safe month.
Speaker 1:We'll see you next month. Bye-bye and don't forget. This show is brought to you by Clearstream Banking, one of the major sponsors of the GFF Summit each year in Luxembourg, and features members of the Clearstream team and special guests expressing their personal opinions, not the opinions of Clearstream as an organisation. And, of course, don't forget that none of the information in this podcast should be taken as legal, tax or other professional advice. See you next time.